International Geothermal Conference, Reykjavík, Sept.

2003 Session #7

Geothermal Plant Design Optimization By Genetic

Digitally signed
by Mohamad Algorithms
Mohamad Ridwan Surono
DN:

Corneliu Popescu, Daniela E. Popescu

cn=Mohamad
Ridwan

Ridwan Surono,
o=POLBAN,
ou=JTK, c=ID
University of Oradea, Str. Armatei Romane nr.5, Oradea, România
Date:
Signatu
re Not
Verified
Surono 2007.07.02
10:34:02 +
07'00'
corneliu_popescu@uoradea.ro, depopescu@uoradea.ro

Abstract
In this paper, it is proposed a genetic algorithms procedure for solving optimal
geothermal plant design where choices on the type of components to be used and
their assembly configuration are driven by reliability objective with the economic
costs associated to the design implementation, system construction and future
operation. We also present the result of our implementation.
Keywords: Genetic algorithms, system design optimization, systems reliability

1 Introduction
When designing a system, several choices must be made concerning the type
of components to be used and their assembly configuration. The choice is driven by
the interaction of reliability/availability objectives with the economic costs associated
to the design implementation, system construction and future operation “Marseguerra
et al.(2000)” ”Fyffe (1968)” “Goldberg (1989)”. Optimization approaches to
determine optimal solutions to design problems have included gradient methods,
dynamic programming, integer programming, mixed integer and nonlinear
programming, and heuristics.
Genetic algorithms are computational tools founded on a direct analogy with the
physical evolution of species and capable of exploring the search space in a very
efficient manner. They have been used to solve several engineering problems and are
particularly effective for combinatorial optimization problems with large, complex
search spaces. Within the reliability field, however, there have been very few
examples of their use.
In our work, the objective function used to measure the fitness of a proposed
solution is the reliability function. Mathematically, then, the problem becomes a
search in the system configuration space of that design which maximizes the value of
the objective function.

2 Generation of time to failure

At the design stage, analyses are to be performed in order to guide the designer
choices in consideration of the many practical aspects which come into play and
which typically generate a conflict between safety requirements and economic needs.
This renders the design effort an optimization one aiming at finding the best
compromise solution.
The geothermal power plant is a component of the cascaded geothermal energy
utilization system, and is used to convert the energy of the geothermal water into
electrical energy using CO2 as working fluid. The elements of the power plant are the
following: heat exchangers to vaporize and condense the CO2, a reciprocating engine
connected with the electric generator, a make-up and expansion CO2 tank, and a CO2
pump.
A good functioning of the power plant following the required thermodynamic
cycle has to insure the heat transfer between the CO2 and the geothermal water or the

S07 Paper048 Page 13

International Geothermal Conference, Reykjavík, Sept. 2003 Session #7

cold water. The control has to maintain constant the CO2 pressure and temperature in
all the important states of the thermodynamic cycle. Together with other specialists,
we decided that we have to implement loops to control the following parameters: t1
(CO2 temperature after vaporization in the heat exchangers), t3 (CO2 temperature after
the condensation in the heat exchanger), and h (level of the liquid CO2 in the tank).
Figure 1 shows the power plant layout, together with the control loops shown using
dotted lines. The reliability model of this structure is given in Figure 2.2.
In the block scheme presented in Figure 2.1 we assume that the vaporizers and
condensers form a series-parallel reliability connection, connectors system is a series
reliability connection and motor, generator, CO2 pump and motor for CO2 pump are
in 2 out of 3 (or 3 out of 4) connection. We analyzed the system considering that the
vaporizers system contains 30 vaporizers.

1
3

2
4

L egend
5
geotherm al w ater
cold w ater
C O 2 circuit
1 vaporizers
2 m otor
7 3 generator
4 C O 2 pum p
5 m otor for C O 2 pum p
6 cond ensers
7 C O 2 tank

Figure 2.1. Geothermal power plant block scheme

1 2 3 6 7 5 4 C onnectors

1. vaporizers
2. motor
3. generator
4. CO2 pump
5. motor for CO2 pump
6. condensors
7. CO2 tank
Figure 2.2 The RBD for the geothermal plant

3 The General Problem For Optimizing A System

We consider a system made up of a series of n nodes, each one performing a given
function. The task of the designer is that of selecting the configuration of each node.
This may be done in several ways, e.g. by choosing different series/parallel
configurations with components of different failure characteristics and therefore of

S07 Paper048 Page 14

International Geothermal Conference, Reykjavík, Sept. 2003 Session #7

different costs “Nakagawa (1981)”. The safety vs. economics conflict rises naturally
as follows:
• Choice of components: choosing the most reliable ones leads to a safe and
high-availability design but it may be largely non-economic due to excessive
component purchase costs; on the other hand, less reliable components provide for
lower purchase costs but loose availability and may increase the risk of costly
accidents.
• Choice of redundancy configuration: highly redundant configurations, with
active or standby components, guarantee high system availability but suffer from
large purchase costs (and perhaps even significant repair costs, if low reliability
units are used); obviously, for assigned component failure and repair
characteristics, low redundancies are economic from the point of view of purchase
costs but weaken the system availability, thus increasing the risk of significant
accidents and the system stoppage time.
In order to find a solution for system optimization, let consider a system with
n components (each one performing a given function) connected as a series reliability
connection “Vladutiu (1989)”. The components are characterized by their fault
probability: q1, q2 , ... qi , ... qn and by their costs: c1, c2 , ... ci , ... cn (Figure 3.1)
q1 q2 qi qn

c1 c2 ci cn Fig.3.a.

q 11 c1 q 21 c2 q i1 ci q n1 cn

q 12 c1 q 22 c2 q i2 ci q n2 cn

q 1m1 c1 q 2m2 c 2 q imi ci q nmn cn

Fig.3.b.

Figure 3.1 The general reliability model for the n components

The reliability, Pp , and cost, C p , functions for this system, are:

n
Pp = ∏ (1 − q )
i =1
1 (3.1)
n
Cp = ∑c i =1
i (3.2)

Each group element will be reserved by a number of identical components

( m1, m2 , ... mi , ... mn ) connected as a parallel reliability connection (Figure 3.b.). We
consider the situation when the groups’ elements are reliability identical, that is:
qi1 = qi 2 = ..... = qimi = qi , where i = 1, n (3.3)
For our system, PD , the probability of functioning without faults, is a monotonously
ascending function. It has n variables: m1, m2 , ... mi , ... mn , which are in the following
relation:
n

∑m ⋅c
i =1
i i = C DM (3.4)

So, mn can be expressed by:

S07 Paper048 Page 15

International Geothermal Conference, Reykjavík, Sept. 2003 Session #7

n −1
C DM − ∑m ⋅c
i =1
i i
mn = (3.5)
cn
and it results for PD :
n −1

n −1
CDM − ∑ mi ⋅ci
∑q
i =1
mi
PD = 1 − i − qn cn (3.6)
i =1

In itia l d a ta : 1.
q 1 , q 2 , ... q i, ... q n
c 1 , c 2 , ... c i, ... c n
C DM G A fo r o b ta in in g th e
o p tim a l s o lu tio n
i = 1

C a lc u la te : D isp lay P D 0 *,
i = i+ 1 ci
βi = C D*
ln q i

no
i = n
STOP

n
CDM − ∑β i ⋅ ln β i
i=1
ln α = n

∑β i
i=1

i = 1

ln α + ln β i
C a lc u la te : ] m i [ = i = i+ 1
ln q i

no
i = n

1.

Figure 3.2 The organigram for the calculus of mi

We have to find the maximum value for the function PD (m1, m2 , ... mi , ... mn ) . So,
∂PD c ⋅ ln qn
= −qimi ⋅ ln qi + q mn i =0 (3.7)
∂mi n
cn
qi ⋅ ln qi qn mn ⋅ ln qn
⇒ = =α (3.8)
ci cn
Let
ci
βi = (3.9)
ln qi
qi mi
⇒α= (3.10)
βi
So,
ln α ⋅ β i β i ⋅ ln α ⋅ β i
mi = = (3.11)
ln qi ci
It results for CDM:
n n n
C DM = ∑ i =1
β i ⋅ ln α ⋅ β i = ∑
i =1
β i ⋅ ln α + ∑ β ⋅ ln β
i =1
i i (3.12)

and

S07 Paper048 Page 16

International Geothermal Conference, Reykjavík, Sept. 2003 Session #7

n
C DM − ∑ β ⋅ ln β
i =1
i i
ln α = n
(3.13)
∑β
i =1
i

It results for mi:

ln α + ln β i
mi = (3.14)
ln qi
Based on relations (3.8), (3.9), (3.13) and (3.14) which are dependent only on the
initial data, we obtain the reliability function, PD 0 , for such a system as being:
n
PDO = 1 − α ⋅ ∑β
i =1
i (3.15)

The problem is that the mi values are real numbers and they must be integers. So,
they must be rounded to: (m1*, m2*, ... mi*, ... mNc*) by respecting the maximum value
⎛ n ⎞
for the CdM ⎜⎜ ∑ mi* ⋅ ci ⎟⎟ when the reliability function is maximized.
⎝ i =1 ⎠
In Figure 3.2 we have an organigram for calculating the mi , i = (1, n ) , values for an
optimal static distributed redundancy.
By an iterative genetic algorithm procedure we try to obtain a maximum value for
the reliability function PD*0 , when C D* ≤ C DM , by selecting k from n mi values with the
]mi [ value, respectively (n-k) mj ( j ≠ i ) values with the m j + 1 value.
For decision-making purposes, the designer defines an objective function, which
accounts for all the relevant aspects. Here we consider as objective function the
reliability for the entire system.
We assume that after an accident the danger management system cannot be
repaired and must be shut down.
For each node a pool of possible configurations is available: the problem is, then,
that of choosing a system assembly by selecting one configuration for each node, with
the aim of maximizing the objective function.

4 The Genetic Algorithm Optimization Approach

The primary target of genetic algorithms is the optimization of an assigned
objective function (fitness) “Painton (1995)”.
A population of 100 chromosomes (bit-strings), each representing a possible
solution to the problem, is initially considered. This population, then, evolves as
dictated by the four fundamental operations of parent’s selection, crossover,
replacement and mutation, for 100 generations.
Using the roulette-wheel selection, also called stochastic sampling with
replacement, performs the selection phase in this work. This is a stochastic algorithm
and involves the following technique: The individuals are mapped to contiguous
segments of a line, such that each individual's segment is equal in size to its fitness. A
random number is generated and the individual whose segment spans the random
number is selected. The process is repeated until the desired number of individuals is
obtained (called mating population). This technique is analogous to a roulette wheel
with each slice proportional in size to the fitness.
The crossover operator used is single-point crossover: one crossover position
k[1,2,...,n-1] – where n is the number of variables of an individual- is selected

S07 Paper048 Page 17

International Geothermal Conference, Reykjavík, Sept. 2003 Session #7

uniformly at random and the variables exchanged between the individuals about this
point, then two new offspring are produced.
For binary valued individuals, mutation means flipping of variable values. For every
individual the variable value to change is chosen uniform at random with a probability
of 10-3.
During the population evolution we eliminate those chromosomes, which encode
infeasible solutions because they violate the cost constraint. With the assigned rules,
which mimic natural selection, the successive generations tend to contain
chromosomes with larger fitness values until a near optimal solution is attained.
Recalling that our system is made up of n nodes, we identify the possible
configurations of each node by a binary value so that the system configuration is
identified by a sequence of n binary numbers, each one indicating a possible node
configuration. For the coding, we choose to take a chromosome made up of a single
gene containing all the values of the node configurations in a string of n bits.
For example, for a node i we can have either a “1” value, when the number of
components for that node is ]mi [ + 1 , either a “0” value, when the number of
components for node i is ]mi [ .
The choice of this coding strategy, as compared to a coding with one gene
dedicated to each node, is such that the crossover generates children-chromosomes
with nodes all equals to the parents except for the one in which the splice occurs. This
avoids excessive dispersion of the genetic patrimony thus favouring convergence.

5 Numerical Applications
The genetic algorithm procedure has been applied to the geothermal simple
system. Given the relative small number of solutions to be spanned in this case, the
best configuration was found also by inspection.
The results thereby obtained were compared to those obtained by the genetic
algorithm and confirmed the good performance of the methodology implemented.
Our genetic algorithm considers a population of chromosomes, each one encoding
a different alternative design solution. For a given design solution, the system
performance over a specified mission time is evaluated in terms of a pre-defined
reliability function. This latter constitutes the objective function to be maximized by
the genetic algorithm through the evolution of the successive generations of the
population in conditions of not overlapping a cost constraint for the system.

Table 4.1
Compone Purchase cost Ci Failure rate λi [10-3
nt i [103 $] y-1]
1 67.5 4.8
2 54 4,3
3 81 4,6
4 45 3.6
5 85.5 3.6
6 58.5 3.7
7 13.5 3.8
8 45 4
The system here considered consists of n = 7 nodes. In TABLE 4.1 we give the
failure rates and the costs for the system components. The maximum cost allowed for
the system is: 1 000 000 $.

S07 Paper048 Page 18

International Geothermal Conference, Reykjavík, Sept. 2003 Session #7

We considered 75 generations for a population of 100 chromosomes and the

evolution was made with a probability for crossovers set as pc = 0,25 and the
probability for the simple mutation set as p m = 0,01 , so, on average, 1% of total bit of
population would undergo mutation.
Figure 4.1 reports the schematic for the optimal configuration found by the genetic
algorithm procedure, which converges only after a few iterations. The reliability value
obtained is 97,435% with a total cost of 995 000$.

Figure 4.1. The optimal configuration

The following simplifying assumptions are made: i) all components have

exponentially distributed failure times; ii) all components a of node. A are equal; iii)
no repair is allowed;
The simple case considered here has allowed us to compute the objective function
analytically and the genetic algorithm was able to converge very rapidly, in a few
iterations. However, for more realistic models we can use a Monte Carlo method for
its evaluation.
It is important to be noted that the optimizing approach presented in this paper can
be extended even for the situation of geothermal plant that includes k-out-of-n: G
schemes (used for reserving the control unit).
The reliability, that is the percentage of successful runs recorded in the simulation
of the resulted system was calculated in EXCEL by using the AVERAGE function
applied to the columns where were recorded the results of individual Monte Carlo
runs. In our case, the resulting reliability was: 97,435%.

6 Conclusion
The genetic algorithm procedure has been applied to a simple system. Given the
relative small number of solutions to be spanned in this case, the best configuration
was found also by inspection. The results thereby obtained were compared to those
obtained by the genetic algorithm and confirmed the good performance of the
methodology implemented.
In conclusion, genetic algorithms can be very useful in solving complex design
problems. The simple case considered here has allowed us to compute the objective
function analytically and the genetic algorithm was able to converge very rapidly, in a
few iterations. However, for more realistic models we can use a Monte Carlo method
for its evaluation.
It is important to be noted that the optimizing approach presented in this paper can
be extended even for the situation of danger control systems that includes k-out-of-n:
G schemes (used for reserving the control unit).

S07 Paper048 Page 19

International Geothermal Conference, Reykjavík, Sept. 2003 Session #7

References

Marseguerra, M., Zio, E., 2000, System design optimization by Genetic Algorithms,
Annual Relliability and Maintainability Symposium, Proceedings`, pp.222-227
Vlăduţiu, M., 1989, Tehnologie de ramură şi fiabilitate, curs litografiat IPTVT,
pp.95-110
Tillman, F. A., Hwang C. L., Kuo, W., 1977, Optimization Techniques for System
Reliability with Redundancy : A Review, IEEE Trans. Reliability, vol. R-26, Aug, pp
148-155.
Fyffe, D.E., Hines, W. W., Lee, N. K., 1981, System Reliability Allocation and a
Computational Algorithm, IEEE Trans. Reliability, Vol R-17, 1968 Jun, pp 64-69.
Nakagawa, Y., Miyazaki, S., Surrogate Constraints Algorithm for Reliability
Optimization Problems with Two Constraints, IEEE Trans. Reliability, Vol R-30, Jun
, pp 175-180.
Goldberg, D. E., 1989, Genetic Algorithms in Search, Optimization, and Machine
Learning,; Addison-Wesley Publishing Company, pp.215-220
Painton, L., Campbell, J., 1995, Genetic Algorithms in Optimization of system
Reliability, IEEE Trans. On Reliab., Vol 44, N. 2, pp 172-178.

S07 Paper048 Page 20